资源描述：
《An Online Learned CRF Model for Multi-Target Tracking》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、AnOnlineLearnedCRFModelforMulti-TargetTrackingBoYangandRamNevatiaInstituteforRoboticsandIntelligentSystems,UniversityofSouthernCaliforniaLosAngeles,CA90089,USA{yangbo
2、nevatia}@usc.eduAbstractWeintroduceanonlinelearningapproachformulti-targettracking.Detectionresponsesareg
3、raduallyassoci-atedintotrackletsinmultiplelevelstoproducefinaltracks.Unlikemostpreviousapproacheswhichonlyfocusonpro-ducingdiscriminativemotionandappearancemodelsforalltargets,wefurtherconsiderdiscriminativefeaturesfordistinguishingdifficultpairsoftargets.Thetrackingprob-le
4、misformulatedusinganonlinelearnedCRFmodel,andistransformedintoanenergyminimizationproblem.Theenergyfunctionsincludeasetofunaryfunctionsthatarebasedonmotionandappearancemodelsfordiscriminat-ingalltargets,aswellasasetofpairwisefunctionsthatarebasedonmodelsfordifferentiating
5、correspondingpairsoftracklets.TheonlineCRFapproachismorepowerfulatFigure1.Examplesoftrackingresultsbyourapproach.distinguishingspatiallyclosetargetswithsimilarappear-ances,aswellasindealingwithcameramotions.Aneffi-field(CRF)modeltobetterdiscriminatingdifferenttargets,cienta
6、lgorithmisintroducedforfindinganassociationwithespeciallydifficultpairs,whicharespatiallyneartargetslowenergycost.Weevaluateourapproachonthreepub-withsimilarappearance.Figure1showssometrackingex-licdatasets,andshowsignificantimprovementscomparedamplesbyourapproach.withsevera
7、lstate-of-artmethods.Toidentifyeachtarget,motionandappearanceinforma-tionareoftenadoptedtoproducediscriminativedescriptors.1.IntroductionMotiondescriptorsareoftenbasedonspeedsanddistancesTrackingmultipletargetsisanimportantbutdifficultbetweentrackletpairs,whileappearancede
8、scriptorsareof-problemincomputervision.Itaimsatfindingtrajectoriestenbasedonglobalorpartbasedcolorhistogramstodistin-ofalltargetswhilemaintainingtheiridentities.Duetogreatguishdifferenttargets.improvementsonobjectdetection,associationbasedtrack-Inmostpreviousassociationbas
9、edtrackingwork,ap-ingapproacheshavebeenproposed[11,17,2,12,18].Theypearancemodelsarepre-defined[1
